| Masque (NarrativeQA only) | 48.7 | 20.98 | 21.95 | 54.74 | Multi-style Generative Reading Comprehension | - |
| BERT-QA with Hard EM objective | - | - | - | 58.8 | A Discrete Hard EM Approach for Weakly Supervised Question Answering | |
| Masque (NarrativeQA + MS MARCO) | 54.11 | 30.43 | 26.13 | 59.87 | Multi-style Generative Reading Comprehension | - |
| Oracle IR Models | 54.60/55.55 | 26.71/27.78 | - | - | The NarrativeQA Reading Comprehension Challenge | |